Frequently asked questions about Responsible AI for Copilot in Viva Pulse

What is Microsoft 365 Copilot in Viva Pulse?

Microsoft 365 Copilot in Viva Pulse uses large language model (LLM) technology to generate a summary of a Viva Pulse report using the report details such as average scores, multiple choice option selections, and open text responses. The summary is included in the email and Teams notifications when the report is generated so that the Pulse author can quickly understand their results at a high level. Authors can share the summary with their team members, demonstrating acknowledgement of responses and transparency.

How is the performance of Copilot in Viva Pulse evaluated?

Performance of Copilot in Viva Pulse is evaluated using these key metrics:

• Precision and Recall: These metrics are crucial for evaluating the quality of a summary generated by Copilot in Viva Pulse. Precision quantifies how many of the generated summaries are relevant, while recall determines how many of the relevant summaries were retrieved.

• User Satisfaction: To gauge user satisfaction, user survey feedback is collected to assess how satisfied users are with the generated summaries.

• Generalizability: To assess how well results are generalized across different use cases, Copilot in Viva Pulse is tested on a diverse set of data and tasks. This involves evaluating system performance on a range of scenarios and domains that were not part of the initial training data.

To help measure initial performance, external experts and testers were invited to find vulnerabilities or biases in the system. The system was carefully tested using Microsoft's Azure AI evaluation tools by intentionally adding potential risks and security loopholes to see how well it handles them. This process helped identify potential issues and improve the system's robustness. The evaluation process is ongoing, with continuous updates and improvements based on user feedback to ensure the accuracy, fairness, and generalizability of Copilot in Viva Pulse.

What are the limitations of Copilot in Viva Pulse? How can users minimize the impact of these limitations when using the system?

The underlying model powering the Copilot in Viva Pulse report summary is trained using the GPT‑4o model. A limitation is that it may not provide a relevant report summary if the report contains unreadable or illegible responses. To mitigate this issue, make sure survey responses are clearly written and understandable.

What operational factors and settings allow for effective and responsible use of Copilot in Viva Pulse?

Copilot in Pulse is designed with a robust filter system that proactively blocks offensive language and prevents the generation of suggestions in sensitive contexts. It is continually improved to detect, address, and remove offensive content generated by Copilot in Viva Pulse, including biased, discriminatory, or abusive outputs.

You are encouraged to report any offensive suggestions you encounter while using Copilot in Viva Pulse. 

Learn more:

Copilot in Viva ​​​​​​​ ​​​​​​​

View and share your Pulse feedback results

Report summary using Copilot in Viva Pulse

Granular access controls for Viva Pulse (for admins)